/*      $NetBSD: subr_percpu.c,v 1.25 2020/05/11 21:37:31 riastradh Exp $       */

/*-
* Copyright (c)2007,2008 YAMAMOTO Takashi,
* All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
* 1. Redistributions of source code must retain the above copyright
*    notice, this list of conditions and the following disclaimer.
* 2. Redistributions in binary form must reproduce the above copyright
*    notice, this list of conditions and the following disclaimer in the
*    documentation and/or other materials provided with the distribution.
*
* THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
* ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
* IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
* ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
* FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
* DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
* OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
* HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
* LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
* OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
* SUCH DAMAGE.
*/

/*
* per-cpu storage.
*/

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: subr_percpu.c,v 1.25 2020/05/11 21:37:31 riastradh Exp $");

#include <sys/param.h>
#include <sys/cpu.h>
#include <sys/kernel.h>
#include <sys/kmem.h>
#include <sys/mutex.h>
#include <sys/percpu.h>
#include <sys/rwlock.h>
#include <sys/vmem.h>
#include <sys/xcall.h>

#define PERCPU_QUANTUM_SIZE     (ALIGNBYTES + 1)
#define PERCPU_QCACHE_MAX       0
#define PERCPU_IMPORT_SIZE      2048

struct percpu {
       unsigned                pc_offset;
       size_t                  pc_size;
       percpu_callback_t       pc_ctor;
       percpu_callback_t       pc_dtor;
       void                    *pc_cookie;
       LIST_ENTRY(percpu)      pc_list;
};

static krwlock_t        percpu_swap_lock        __cacheline_aligned;
static vmem_t *         percpu_offset_arena     __read_mostly;
static struct {
       kmutex_t        lock;
       unsigned int    nextoff;
       LIST_HEAD(, percpu) ctor_list;
       struct lwp      *busy;
       kcondvar_t      cv;
} percpu_allocation __cacheline_aligned;

static percpu_cpu_t *
cpu_percpu(struct cpu_info *ci)
{

       return &ci->ci_data.cpu_percpu;
}

static unsigned int
percpu_offset(percpu_t *pc)
{
       const unsigned int off = pc->pc_offset;

       KASSERT(off < percpu_allocation.nextoff);
       return off;
}

/*
* percpu_cpu_swap: crosscall handler for percpu_cpu_enlarge
*/
__noubsan
static void
percpu_cpu_swap(void *p1, void *p2)
{
       struct cpu_info * const ci = p1;
       percpu_cpu_t * const newpcc = p2;
       percpu_cpu_t * const pcc = cpu_percpu(ci);

       KASSERT(ci == curcpu() || !mp_online);

       /*
        * swap *pcc and *newpcc unless anyone has beaten us.
        */
       rw_enter(&percpu_swap_lock, RW_WRITER);
       if (newpcc->pcc_size > pcc->pcc_size) {
               percpu_cpu_t tmp;
               int s;

               tmp = *pcc;

               /*
                * block interrupts so that we don't lose their modifications.
                */

               s = splhigh();

               /*
                * copy data to new storage.
                */

               memcpy(newpcc->pcc_data, pcc->pcc_data, pcc->pcc_size);

               /*
                * this assignment needs to be atomic for percpu_getptr_remote.
                */

               pcc->pcc_data = newpcc->pcc_data;

               splx(s);

               pcc->pcc_size = newpcc->pcc_size;
               *newpcc = tmp;
       }
       rw_exit(&percpu_swap_lock);
}

/*
* percpu_cpu_enlarge: ensure that percpu_cpu_t of each cpus have enough space
*/

static void
percpu_cpu_enlarge(size_t size)
{
       CPU_INFO_ITERATOR cii;
       struct cpu_info *ci;

       for (CPU_INFO_FOREACH(cii, ci)) {
               percpu_cpu_t pcc;

               pcc.pcc_data = kmem_alloc(size, KM_SLEEP); /* XXX cacheline */
               pcc.pcc_size = size;
               if (!mp_online) {
                       percpu_cpu_swap(ci, &pcc);
               } else {
                       uint64_t where;

                       where = xc_unicast(0, percpu_cpu_swap, ci, &pcc, ci);
                       xc_wait(where);
               }
               KASSERT(pcc.pcc_size <= size);
               if (pcc.pcc_data != NULL) {
                       kmem_free(pcc.pcc_data, pcc.pcc_size);
               }
       }
}

/*
* percpu_backend_alloc: vmem import callback for percpu_offset_arena
*/

static int
percpu_backend_alloc(vmem_t *dummy, vmem_size_t size, vmem_size_t *resultsize,
   vm_flag_t vmflags, vmem_addr_t *addrp)
{
       unsigned int offset;
       unsigned int nextoff;

       ASSERT_SLEEPABLE();
       KASSERT(dummy == NULL);

       if ((vmflags & VM_NOSLEEP) != 0)
               return ENOMEM;

       size = roundup(size, PERCPU_IMPORT_SIZE);
       mutex_enter(&percpu_allocation.lock);
       offset = percpu_allocation.nextoff;
       percpu_allocation.nextoff = nextoff = percpu_allocation.nextoff + size;
       mutex_exit(&percpu_allocation.lock);

       percpu_cpu_enlarge(nextoff);

       *resultsize = size;
       *addrp = (vmem_addr_t)offset;
       return 0;
}

static void
percpu_zero_cb(void *vp, void *vp2, struct cpu_info *ci)
{
       size_t sz = (uintptr_t)vp2;

       memset(vp, 0, sz);
}

/*
* percpu_zero: initialize percpu storage with zero.
*/

static void
percpu_zero(percpu_t *pc, size_t sz)
{

       percpu_foreach(pc, percpu_zero_cb, (void *)(uintptr_t)sz);
}

/*
* percpu_init: subsystem initialization
*/

void
percpu_init(void)
{

       ASSERT_SLEEPABLE();
       rw_init(&percpu_swap_lock);
       mutex_init(&percpu_allocation.lock, MUTEX_DEFAULT, IPL_NONE);
       percpu_allocation.nextoff = PERCPU_QUANTUM_SIZE;
       LIST_INIT(&percpu_allocation.ctor_list);
       percpu_allocation.busy = NULL;
       cv_init(&percpu_allocation.cv, "percpu");

       percpu_offset_arena = vmem_xcreate("percpu", 0, 0, PERCPU_QUANTUM_SIZE,
           percpu_backend_alloc, NULL, NULL, PERCPU_QCACHE_MAX, VM_SLEEP,
           IPL_NONE);
}

/*
* percpu_init_cpu: cpu initialization
*
* => should be called before the cpu appears on the list for CPU_INFO_FOREACH.
* => may be called for static CPUs afterward (typically just primary CPU)
*/

void
percpu_init_cpu(struct cpu_info *ci)
{
       percpu_cpu_t * const pcc = cpu_percpu(ci);
       struct percpu *pc;
       size_t size = percpu_allocation.nextoff; /* XXX racy */

       ASSERT_SLEEPABLE();

       /*
        * For the primary CPU, prior percpu_create may have already
        * triggered allocation, so there's nothing more for us to do
        * here.
        */
       if (pcc->pcc_size)
               return;
       KASSERT(pcc->pcc_data == NULL);

       /*
        * Otherwise, allocate storage and, while the constructor list
        * is locked, run constructors for all percpus on this CPU.
        */
       pcc->pcc_size = size;
       if (size) {
               pcc->pcc_data = kmem_zalloc(pcc->pcc_size, KM_SLEEP);
               mutex_enter(&percpu_allocation.lock);
               while (percpu_allocation.busy)
                       cv_wait(&percpu_allocation.cv,
                           &percpu_allocation.lock);
               percpu_allocation.busy = curlwp;
               LIST_FOREACH(pc, &percpu_allocation.ctor_list, pc_list) {
                       KASSERT(pc->pc_ctor);
                       mutex_exit(&percpu_allocation.lock);
                       (*pc->pc_ctor)((char *)pcc->pcc_data + pc->pc_offset,
                           pc->pc_cookie, ci);
                       mutex_enter(&percpu_allocation.lock);
               }
               KASSERT(percpu_allocation.busy == curlwp);
               percpu_allocation.busy = NULL;
               cv_broadcast(&percpu_allocation.cv);
               mutex_exit(&percpu_allocation.lock);
       }
}

/*
* percpu_alloc: allocate percpu storage
*
* => called in thread context.
* => considered as an expensive and rare operation.
* => allocated storage is initialized with zeros.
*/

percpu_t *
percpu_alloc(size_t size)
{

       return percpu_create(size, NULL, NULL, NULL);
}

/*
* percpu_create: allocate percpu storage and associate ctor/dtor with it
*
* => called in thread context.
* => considered as an expensive and rare operation.
* => allocated storage is initialized by ctor, or zeros if ctor is null
* => percpu_free will call dtor first, if dtor is nonnull
* => ctor or dtor may sleep, even on allocation
*/

percpu_t *
percpu_create(size_t size, percpu_callback_t ctor, percpu_callback_t dtor,
   void *cookie)
{
       vmem_addr_t offset;
       percpu_t *pc;

       ASSERT_SLEEPABLE();
       (void)vmem_alloc(percpu_offset_arena, size, VM_SLEEP | VM_BESTFIT,
           &offset);

       pc = kmem_alloc(sizeof(*pc), KM_SLEEP);
       pc->pc_offset = offset;
       pc->pc_size = size;
       pc->pc_ctor = ctor;
       pc->pc_dtor = dtor;
       pc->pc_cookie = cookie;

       if (ctor) {
               CPU_INFO_ITERATOR cii;
               struct cpu_info *ci;
               void *buf;

               /*
                * Wait until nobody is using the list of percpus with
                * constructors.
                */
               mutex_enter(&percpu_allocation.lock);
               while (percpu_allocation.busy)
                       cv_wait(&percpu_allocation.cv,
                           &percpu_allocation.lock);
               percpu_allocation.busy = curlwp;
               mutex_exit(&percpu_allocation.lock);

               /*
                * Run the constructor for all CPUs.  We use a
                * temporary buffer wo that we need not hold the
                * percpu_swap_lock while running the constructor.
                */
               buf = kmem_alloc(size, KM_SLEEP);
               for (CPU_INFO_FOREACH(cii, ci)) {
                       memset(buf, 0, size);
                       (*ctor)(buf, cookie, ci);
                       percpu_traverse_enter();
                       memcpy(percpu_getptr_remote(pc, ci), buf, size);
                       percpu_traverse_exit();
               }
               explicit_memset(buf, 0, size);
               kmem_free(buf, size);

               /*
                * Insert the percpu into the list of percpus with
                * constructors.  We are now done using the list, so it
                * is safe for concurrent percpu_create or concurrent
                * percpu_init_cpu to run.
                */
               mutex_enter(&percpu_allocation.lock);
               KASSERT(percpu_allocation.busy == curlwp);
               percpu_allocation.busy = NULL;
               cv_broadcast(&percpu_allocation.cv);
               LIST_INSERT_HEAD(&percpu_allocation.ctor_list, pc, pc_list);
               mutex_exit(&percpu_allocation.lock);
       } else {
               percpu_zero(pc, size);
       }

       return pc;
}

/*
* percpu_free: free percpu storage
*
* => called in thread context.
* => considered as an expensive and rare operation.
*/

void
percpu_free(percpu_t *pc, size_t size)
{

       ASSERT_SLEEPABLE();
       KASSERT(size == pc->pc_size);

       /*
        * If there's a constructor, take the percpu off the list of
        * percpus with constructors, but first wait until nobody is
        * using the list.
        */
       if (pc->pc_ctor) {
               mutex_enter(&percpu_allocation.lock);
               while (percpu_allocation.busy)
                       cv_wait(&percpu_allocation.cv,
                           &percpu_allocation.lock);
               LIST_REMOVE(pc, pc_list);
               mutex_exit(&percpu_allocation.lock);
       }

       /* If there's a destructor, run it now for all CPUs.  */
       if (pc->pc_dtor) {
               CPU_INFO_ITERATOR cii;
               struct cpu_info *ci;
               void *buf;

               buf = kmem_alloc(size, KM_SLEEP);
               for (CPU_INFO_FOREACH(cii, ci)) {
                       percpu_traverse_enter();
                       memcpy(buf, percpu_getptr_remote(pc, ci), size);
                       explicit_memset(percpu_getptr_remote(pc, ci), 0, size);
                       percpu_traverse_exit();
                       (*pc->pc_dtor)(buf, pc->pc_cookie, ci);
               }
               explicit_memset(buf, 0, size);
               kmem_free(buf, size);
       }

       vmem_free(percpu_offset_arena, (vmem_addr_t)percpu_offset(pc), size);
       kmem_free(pc, sizeof(*pc));
}

/*
* percpu_getref:
*
* => safe to be used in either thread or interrupt context
* => disables preemption; must be bracketed with a percpu_putref()
*/

void *
percpu_getref(percpu_t *pc)
{

       kpreempt_disable();
       return percpu_getptr_remote(pc, curcpu());
}

/*
* percpu_putref:
*
* => drops the preemption-disabled count after caller is done with per-cpu
*    data
*/

void
percpu_putref(percpu_t *pc)
{

       kpreempt_enable();
}

/*
* percpu_traverse_enter, percpu_traverse_exit, percpu_getptr_remote:
* helpers to access remote cpu's percpu data.
*
* => called in thread context.
* => percpu_traverse_enter can block low-priority xcalls.
* => typical usage would be:
*
*      sum = 0;
*      percpu_traverse_enter();
*      for (CPU_INFO_FOREACH(cii, ci)) {
*              unsigned int *p = percpu_getptr_remote(pc, ci);
*              sum += *p;
*      }
*      percpu_traverse_exit();
*/

void
percpu_traverse_enter(void)
{

       ASSERT_SLEEPABLE();
       rw_enter(&percpu_swap_lock, RW_READER);
}

void
percpu_traverse_exit(void)
{

       rw_exit(&percpu_swap_lock);
}

void *
percpu_getptr_remote(percpu_t *pc, struct cpu_info *ci)
{

       return &((char *)cpu_percpu(ci)->pcc_data)[percpu_offset(pc)];
}

/*
* percpu_foreach: call the specified callback function for each cpus.
*
* => must be called from thread context.
* => callback executes on **current** CPU (or, really, arbitrary CPU,
*    in case of preemption)
* => caller should not rely on the cpu iteration order.
* => the callback function should be minimum because it is executed with
*    holding a global lock, which can block low-priority xcalls.
*    eg. it's illegal for a callback function to sleep for memory allocation.
*/
void
percpu_foreach(percpu_t *pc, percpu_callback_t cb, void *arg)
{
       CPU_INFO_ITERATOR cii;
       struct cpu_info *ci;

       percpu_traverse_enter();
       for (CPU_INFO_FOREACH(cii, ci)) {
               (*cb)(percpu_getptr_remote(pc, ci), arg, ci);
       }
       percpu_traverse_exit();
}

struct percpu_xcall_ctx {
       percpu_callback_t  ctx_cb;
       void              *ctx_arg;
};

static void
percpu_xcfunc(void * const v1, void * const v2)
{
       percpu_t * const pc = v1;
       struct percpu_xcall_ctx * const ctx = v2;

       (*ctx->ctx_cb)(percpu_getref(pc), ctx->ctx_arg, curcpu());
       percpu_putref(pc);
}

/*
* percpu_foreach_xcall: call the specified callback function for each
* cpu.  This version uses an xcall to run the callback on each cpu.
*
* => must be called from thread context.
* => callback executes on **remote** CPU in soft-interrupt context
*    (at the specified soft interrupt priority).
* => caller should not rely on the cpu iteration order.
* => the callback function should be minimum because it may be
*    executed in soft-interrupt context.  eg. it's illegal for
*    a callback function to sleep for memory allocation.
*/
void
percpu_foreach_xcall(percpu_t *pc, u_int xcflags, percpu_callback_t cb,
                    void *arg)
{
       struct percpu_xcall_ctx ctx = {
               .ctx_cb = cb,
               .ctx_arg = arg,
       };
       CPU_INFO_ITERATOR cii;
       struct cpu_info *ci;

       for (CPU_INFO_FOREACH(cii, ci)) {
               xc_wait(xc_unicast(xcflags, percpu_xcfunc, pc, &ctx, ci));
       }
}